An AI model combining preoperative MRI and clinical factors predict positive surgical margin risk after radical prostatectomy in 1,177 Prostate Cancer (PrCA) patients, with an external validation AUC of 0.843 across six independent centers. In this retrospective multicenter study, the multimodal deep learning model outperformed an imaging-only model (0.711, p=0.003) and a clinical-only model (0.676), suggesting better pre-op risk stratification for surgical planning.

Why It Matters To Your Practice

• Positive surgical margins can affect prognosis, counseling, and decisions around operative approach.

• A pre-op tool that integrates MRI with routine clinical data could help identify patients at higher margin risk before radical prostatectomy.

• External validation across multiple centers suggests the model may be more generalizable than single-site AI tools.

Clinical Implications

• High-risk predictions could prompt more deliberate nerve-sparing tradeoffs, wider excision planning, or more detailed informed consent.

• The model may help frame discussions about oncologic control versus functional preservation in borderline cases.

• Because the study was retrospective, the tool is not yet ready to replace clinician judgment or established risk assessment workflows.

Insights

• The model used feature disentanglement to reduce hospital-specific variation and multimodal fusion to combine MRI-derived signals with clinical variables.

• Interpretability was supported with gradient-weighted class activation mapping, showing which image regions influenced predictions.

• The strongest performance signal came from the combined model, not MRI alone or clinical factors alone.

The Bottom Line

• This multicenter study suggests AI can improve preoperative prediction of positive surgical margins in PrCA when MRI and clinical data are combined.

• If prospectively validated and integrated into workflow, such models could support more individualized surgical planning before radical prostatectomy.